epic: Group B — trust model extension (capability-scoped + multi-dimensional)

[mdproctor]

Overview

Extends the trust model to support capability-scoped and multi-dimensional scores. Do Group A first, then B.

Parent epic: #48

Issues (B1 before B2; B3 is independent of B1/B2)

Issue	Size	Depends on	Description
Capability-scoped attestations (B1)	S	—	Add capabilityTag field to LedgerAttestation
Capability-scoped trust scores (B2)	M	B1	ActorTrustScore keyed by (actorId, capabilityTag) + TrustScoreJob extension
Multi-dimensional trust infrastructure (B3)	L	—	Dimension-labelled Beta models per actor

What this enables

Without Group B, trust is one number per actor. With Group B:

• Agent X: security-review trust 0.85, architecture-review trust 0.4
• TrustGateService can gate specifically on security-review trust
• The routing strategy can differentiate: "good overall agent but not for security work"

This is the key gap between CaseHub and Gastown's multi-dimensional stamps — CaseHub's model becomes superior because it is computed automatically rather than manually assigned.

Motivating use case

A PR contains both style issues and cryptographic code. The system routes it to two agents: one with high style-review trust (quick assignment), one with high security-review trust (high threshold enforced by TrustGateService). Each agent's outcome feeds back into their respective capability-scoped trust score — not their overall score.